I like The Pool and the one published game based on it, The Questing Beast. It's a neat little free roleplaying game that's inestimably easy to customize. There's quite a tradition of Pool variants - my masterpiece in that field is the unfortunately-named Snowball. Below are some other humble tweaks, less drastic - more variants of the Pool can be found here.
Flipping the Beast
The idea of Anti-Pool proposed by Mike Holmes appeals to me. So does the nature of The Questing Beast, where there are Monologues of Defeat as well as Monologues of Victory. This particular set-up combines the two. It was used in Snowball, but I think it can be applied in a wider set of applications.
In short, you lose dice on an Monologue of Victory, your dice remain the same on a Guided Event, and if you roll a Monologue of Defeat, you earn as many dice as the GM gave you for your action in the first place.
Beast with two Backs
Having flipped the beast, I also crunched the numbers to see how The Beast's setup would work if the penalty was on the Guided Event. It looked pretty good, and on my lazy days I like the idea that the players only pay when the GM does the narration. Using the Beast with two Backs, you lose dice on a Guided Event instead of either Monologue. You gain a die on a Monologue of Defeat.
On the Costs of Traits
In standard Pool, traits cost the bonus multiplied by itself (1...4...9...16). In The Questing Beast, this was changed to twice the bonus (2...4...6...8). Obviously, the latter has a much smaller curve, but it increases the buy-in value, something that I wasn't entirely fond of. The idea that a new, minor (1 die) Trait could be added at pretty much any time appealed to me, and in some ways still does. Some variants of the Pool don't quantify the traits at all, but I like the ranking system - it shows how important the Trait's success is to the player. Other price structures that I've personally used include squares (1...2...4...8) and a pyramidical structure (1...3...6...10).
Although the square cost structure is cheaper than the pyramidical for values below five, the pyramidical's curve is nicer, and it's easier to understand (the underlying formula is "each level costs its value, plus the value of each preceding level" - going from 1 to 2 costs 2 points, 2 to 3 costs 3 more points, etc). This is the reason why Snowball uses it, as well as why I use it in most of my Pool games, Snowball or no.